Find leads
Run log
idle.
Past runs
Loading…
Collect prospects from a CSV
Apollo export or any CSV with email, company,
website (first/last name, title, linkedin optional). Runs email verify,
MX routing, screenshots, and PageSpeed — no AI, no tokens.
Progress
idle.
Send approved drafts
Only *.approved.md send. Per-mailbox daily caps, random
3–12 min delays, and the weekend skip are enforced in code.
Send log
idle.
Log
idle.
Pipeline
Mailbox health
Template performance
Opener styles
Outcome attribution — what converts to meetings & wins, not just replies
By lead source
By angle
Drafting guidance (what the learning loop tells Claude)
loading…
Claude models
All headless, on your subscription — never the API. Drafting is the product, so it gets the strongest model; classification and web lookup are simpler tasks that default to smaller tiers to preserve your usage quota.
Learning (bandit)
Off by default. The Thompson-sampling bandit always records send/reply data, but while this is off it never influences what gets written — drafting runs purely on the playbook + each prospect's research. Turn it on once you have ~a month of real reply data, so it learns from signal instead of noise.
Mailboxes
Add mailbox
Bump sequences
Threaded nudges sent automatically to non-repliers after approval —
{first_name}/{company}/{domain} placeholders work.
Loading…
Google OAuth client
Suppression list
One email or domain per line — existing clients, opt-outs. Suppression is global across campaigns.